tama 0.0.1

Multi-agent AI framework — build, run, and trace agent pipelines from the command line
Documentation

tama 玉

Write skills in Markdown. Brew them into agents. Ship as distroless Docker.

Tama (玉) — a magical jewel, a crystal that holds power. You bring the raw skills. Tama fuses them into something greater.

tama add react                 # install pattern skill
tama add search-web            # install tool skill
tama brew                      # resolve deps → generate Dockerfile
docker build -t my-agent .
docker run --env-file .env my-agent "research quantum computing trends"

Philosophy

  • Skills are Markdown files. Human-readable, git-diffable, writable in any editor.
  • Patterns are skills. chain-of-thought, critic, tree-of-thought — all just skills that compose others. No special syntax.
  • Deps are declared, not installed at runtime. tama brew resolves everything at compile time → distroless image with zero bloat.
  • The runtime is a Rust binary. No bash, no Python orchestrator, no framework. Just a fast binary calling other binaries.

Project layout

my-agent/
  tama.json              ← installed skills (like package.json)
  skills/
    react/SKILL.md
    search-web/SKILL.md
    write-report/SKILL.md
  Dockerfile             ← generated by tama brew, do not edit

Commands

tama add <n>                   # install from registry
tama add github:user/repo      # install from GitHub
tama add ./local-skill         # install local skill
tama list                      # show installed skills + dep tree
tama brew                      # compile → Dockerfile
tama run "your task"           # brew + docker build + docker run
tama serve                     # brew + build + run as HTTP/SSE server

Docker output

Multi-stage distroless build:

debian:bookworm-slim  →  extract binaries/libs  →  gcr.io/distroless/cc-debian12
     (builder)              (only what's needed)         (final ~8MB + deps)

No shell. No package managers. No root. Just your agent.